Kekc/dogsML

Machine Learning pet project

Machine Learning pet project.

Various implementation of the neural nets:

• Simple naive versions built with numpy - source
• Basic Tensorflow version - source
• Convolutional neural net - source
• Transfer learning with Tensorflow MobileNet - source
• Utils to create data pipelines and to use saved tensorflow models

---

Dog's Breed classification model

Model Architecture: Transfer learning with Tensorflow.MobileNetV3Large neural net. Dataset: Stanford Dogs Dataset

The most accurate model achieved 77% validation accuracy:

• Model source
• Google colab
• Model is deployed via the Telegram Bot

Used datasets:

• http://vision.stanford.edu/aditya86/ImageNetDogs/
• https://www.kaggle.com/datasets/prasunroy/natural-images
• https://www.kaggle.com/datasets/jessicali9530/stanford-dogs-dataset

Inspired by: https://www.coursera.org/specializations/deep-learning

---

Run pip install -e . in the ROOT folder for the local development